摘要
纹理图像中的不规则部分通常称为疵点。纹理通常由空间分布和灰度分布共同描述,由于灰度共生矩阵能兼顾二者,因此具有很好的描述纹理的能力,不过其对纹理的正常部分与不正常部分的区分能力仍然有限,且计算效率低。为克服灰度共生矩阵以上的不足,提出了一种用模糊类别共生矩阵的方法来检测不规则纹理。该方法首先通过学习纹理的基本特征,如纹理的灰度概率密度分布、纹理主方向和周期等来确定模糊类别共生矩阵的一些关键参数,并将灰度级划分为几个纹理色调类别;然后根据后验概率函数得出各个灰度级对每个色调类别的模糊隶属度,同时计算模糊类别共生矩阵,并提取一些更为简单的特征;最后利用异常点检测的方法,即可很好地区分正常纹理和疵点。实践证明,该方法不仅比已有的灰度共生矩阵方法更简单,计算效率更高,而且能更好地表示不规则纹理。
In the texture image, the imperfect part is denoted as defect. Texture is usually depicted by a gray-level distribution along with a certain spatial interaction. Thus, gray-level co-occurrence matrix （GLCM） is an appropriate candidate to depict texture because of its capability of blending spatial interaction with gray-level distribution. However, GLCM is considered as deficient in discriminating the normal and abnormal parts of texture, and in computation efficiency as well. In order to overcome these drawbacks of GLCM, a method of fuzzy label co-occurrence matrix（FLCM） is proposed to detect the textural imperfection. In this method, textural features such as the probability density distribution of the gray levels, the intrinsic dominant orientation and periodicity in the texture, are extracted firstly to set some key parameters of FLCM, and then all gray-levels are classified into several textural tonal classes in a certain rule; the fuzzy membership degrees of each gray-level to each tonal class are computed based on the corresponding posteriori probability, finally the FLCMs are calculated and some simple features are extracted from the FLCMs, and outlier detection is applied to discriminate imperfection from normal texture. It is proved practically that this method is simpler and has better performance in detecting textural imperfection than GLCM.
出处
《中国图象图形学报》
CSCD
北大核心
2007年第1期92-97,共6页
Journal of Image and Graphics
关键词
纹理疵点检测
模糊类别共生矩阵
模糊隶属函数
灰度共生矩阵
textural defect detection, fuzzy label co-occurrence matrix（FLCM）, fuzzy membership function, gray-level co-occurrence matrix（ GLCM ）
